Deep learning seismic lithology prediction method based on geological statistical characteristic law constraint

Abstract

The invention relates to a deep learning seismic lithology prediction method based on geostatistical characteristic law constraints. The method comprises the following steps: S1, obtaining multi-seismic information; s2, acquiring geological data and logging data, counting features of the geological data and the logging data, and numeralizing the features to obtain various geological statistical features; s3, constructing a data set with labels required by training; s4, training, verifying and testing the deep convolutional neural network model based on a data set with a label required by training to obtain a deep learning seismic lithology prediction model; and S5, acquiring actual seismic data, and inputting the actual seismic data into the deep learning seismic lithology prediction model to obtain an actual seismic lithology prediction result. Compared with the prior art, the method has the advantages of improving the seismic lithology prediction precision and reliability and the like.

Description

Technical Field

[0001] The present invention relates to the field of seismic exploration, and in particular to a deep learning seismic lithology prediction method based on geological statistical characteristic law constraints. Background Art

[0002] Traditional model-driven pre-stack inversion and post-stack seismic reservoir prediction technologies are limited by seismic resolution and are difficult to effectively predict thin layers. Traditional frequency-division seismic technology can only provide qualitative and semi-quantitative descriptions of thin layers, and lacks the joint use of pre-stack seismic gathers and multiple seismic attributes. Most of these methods rely on physical models, and it is usually difficult to effectively identify complex concealed reservoirs and reservoirs disturbed by coal seams. Although deep learning-based lithology prediction methods introduce the advantage of data-driven, they usually only rely on seismic data and very little logging information, ignoring the comprehensive use of geological prior knowledge. In actual geological exploration, lithology prediction tasks often involve multiple strata, which vary significantly from stratum to stratum, and there are complex changes in the geological statistical characteristics of different strata.

[0003] Existing seismic lithology prediction methods only construct training sample sets based on multiple seismic information, resulting in low prediction accuracy and reliability. Summary of the invention

[0004] The purpose of the present invention is to provide a deep learning seismic lithology prediction method based on geological statistical characteristic law constraints in order to improve the accuracy and reliability of seismic lithology prediction.

[0005] The purpose of the present invention can be achieved by the following technical solutions:

[0006] A deep learning seismic lithology prediction method based on geological statistical characteristic law constraints, the method comprises the following steps:

[0007] S1, obtaining post-stack seismic data and pre-stack gather data, performing feature extraction and seismic inversion on the post-stack seismic data and the pre-stack gather data, and obtaining seismic attributes and elastic parameter inversion results, wherein the seismic attributes, elastic parameter inversion results, post-stack seismic data and pre-stack gather data constitute multi-seismic information;

[0008] S2. Obtain geological data and well logging data, count the features of the geological data and well logging data and digitize the features, and the specific steps of digitizing the features are: uniquely encode each stratigraphic unit to obtain a coding representation, and obtain a priori probability vector of the lithology type of each stratigraphic unit, obtain a conversion probability vector between different lithologies, obtain the thickness of mudstone, sandstone and coal seam of each stratigraphic unit, and obtain a thickness distribution vector, wherein the coding represents the stratigraphic unit, the priori probability vector, the conversion probability vector and the thickness distribution vector as multiple geological statistical features;

[0009] S3. Through the time-depth relationship, multiple seismic information and multiple geological statistical features are associated, corresponding lithology labels are set for the multiple seismic information, and multiple geological statistical features are used to constrain the multiple seismic information to obtain the labeled data set required for training;

[0010] S4. Training, verifying and testing the deep convolutional neural network model based on the labeled data set required for training to obtain a deep learning seismic lithology prediction model;

[0011] S5. Obtain actual seismic data, input the actual seismic data into the deep learning seismic lithology prediction model, and obtain actual seismic lithology prediction results.

[0012] Further, the encoding is expressed as:

[0013] For the 6 stratigraphic units, unit 1 is encoded as [1,0,0,0,0,0], unit 2 is encoded as [0,1,0,0,0,0], unit 3 is encoded as [0,0,1,0,0,0], unit 4 is encoded as [0,0,0,1,0,0], unit 5 is encoded as [0,0,0,0,1,0], and unit 6 is encoded as [0,0,0,0,0,1].

[0014] Furthermore, the prior probability vector of lithology type is [p M ,p S ,p C ], p M ,p S ,p C Represent the prior probabilities of mudstone, sandstone and coal seams respectively.

[0015] Furthermore, the conversion probability vector between different lithologies is:

[0016] [T M,M ,T M,S ,T M,C ,T S,M ,T S,S ,T S,C ,T C,M ,T C,S ,T C,C ]

[0017] Among them, T M,M ,T M,S ,T M,C ,T S,M ,T S,S ,T S,C ,T C,M ,T C,S ,T C,CThey respectively represent the conversion probabilities of mudstone to mudstone, mudstone to sandstone, mudstone to coal seam, sandstone to mudstone, sandstone to sandstone, sandstone to coal seam, coal seam to mudstone, coal seam to sandstone, and coal seam to coal seam.

[0018] Furthermore, the thickness distribution vector is:

[0019]

[0020] in They represent the distribution of mudstone 1-5ms, mudstone 5-10ms, mudstone 10-20ms, mudstone 20-70ms, sandstone 1-5ms, sandstone 5-10ms, sandstone 10-20ms, sandstone 20-70ms, They respectively represent the 1-2ms category distribution of coal seams and the 2-3ms category distribution of coal seams.

[0021] Furthermore, the deep convolutional neural network model is trained in a sequence-to-point manner, and the input label sequence consists of multiple sampling points adjacent to the point to be predicted.

[0022] Furthermore, the deep convolutional neural network model is trained using a cross blind testing method.

[0023] Furthermore, the specific steps of training using the cross blind test method are:

[0024] The labeled data set is divided into training well data and a blind well logging data. The training well data is input into the deep convolutional neural network model for training verification, and the blind well logging data is input into the trained deep convolutional neural network model for testing. The above steps are repeated to obtain the blind test results of each well. Based on the blind test results of each well, the well data with prediction accuracy less than the threshold are screened, and the screened well data are input into the deep convolutional neural network model for training again.

[0025] Furthermore, an early stopping mechanism is adopted in the training process of the deep convolutional neural network model.

[0026] Furthermore, the learning rate is dynamically adjusted during the training of the deep convolutional neural network model.

[0027] Compared with the prior art, the present invention has the following beneficial effects:

[0028] The present invention converts complex geological information into numerical features that can be understood and processed by deep neural networks by encoding and numerically processing four types of geostatistical features (stratigraphic units, lithology prior probability, lithology conversion matrix and lithology thickness distribution). Among them, the stratigraphic unit is represented by a unique hot encoding, and the lithology prior probability, lithology conversion matrix and lithology thickness distribution are processed by a probabilistic numerical representation method, which provides clear constraints for lithology prediction. The four geostatistical features are used as the input of the deep convolutional neural network, so that the model can be effectively constrained by the input geological prior information while using seismic data for lithology prediction. This method enhances the stability and prediction accuracy of the model, while making full use of geological background knowledge, optimizing the lithology prediction process, combining a variety of seismic information with the constraints of a variety of geostatistical features, thereby achieving a more comprehensive and accurate seismic lithology prediction. Through this synergistic effect, the model can more accurately capture complex lithology changes, especially in the prediction of complex geological bodies such as coal seams. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] Figure 1 is a flow chart of the present invention;

[0030] Figure 2 The deep convolutional neural network architecture of multiple seismic information and multiple geological constraints of the present invention;

[0031] Figure 3 This is the cross blind test result of two wells;

[0032] Figure 4 is the prediction result of seismic lithology slice, where Figure 4 (a) is the prediction result using only multi-earthquake information. Figure 4 (b) Prediction results of the synergistic effect of multiple seismic information and multiple geological statistical characteristic constraints. DETAILED DESCRIPTION

[0033] The present invention is described in detail below in conjunction with the accompanying drawings and specific embodiments. This embodiment is implemented based on the technical solution of the present invention, and provides a detailed implementation method and specific operation process, but the protection scope of the present invention is not limited to the following embodiments.

[0034] The present invention proposes a deep learning seismic lithology prediction method based on geological statistical characteristic law constraints. The flowchart of the method is as follows: Figure 1As shown. Taking into account the stratigraphic information, lithology prior probability, lithology succession relationship and lithology thickness distribution, a set of training sample sets with multiple geological statistical features and multiple seismic information was constructed. The blind test results show that the introduction of geological prior constraints can significantly improve the prediction accuracy and reliability of sandstone and thin coal seams. Compared with the method of constructing training sample sets based only on multiple seismic information, the prediction results of this method are more consistent with the actual geological prior cognition in complex reservoirs with large differences in lithology distribution, and the predicted lithology slices are more in line with the sedimentary environment characteristics of each geological unit.

[0035] The method of the present invention comprises the following steps:

[0036] S1, obtaining post-stack seismic data and pre-stack gather data, performing feature extraction and seismic inversion on the post-stack seismic data and pre-stack gather data, obtaining seismic attribute and elastic parameter inversion results, and the post-stack seismic data and pre-stack gather data of seismic attributes and elastic parameters inversion results constitute multi-seismic information;

[0037] S2. Obtain geological data and well logging data, count the features of the geological data and well logging data and digitize the features, and the specific steps of digitizing the features are: uniquely encode each stratigraphic unit to obtain a coding representation, and obtain a priori probability vector of the lithology type of each stratigraphic unit, obtain a conversion probability vector between different lithologies, obtain the thickness of mudstone, sandstone and coal seam of each stratigraphic unit, and obtain a thickness distribution vector, wherein the coding represents the stratigraphic unit, the priori probability vector, the conversion probability vector and the thickness distribution vector as multiple geological statistical features;

[0038] S3. Through the time-depth relationship, multiple seismic information and multiple geological statistical features are associated, corresponding lithology labels are set for the multiple seismic information, and multiple geological statistical features are used to constrain the multiple seismic information to obtain the labeled data set required for training;

[0039] S4. Training, verifying and testing the deep convolutional neural network model based on the labeled data set required for training to obtain a deep learning seismic lithology prediction model;

[0040] S5. Obtain actual seismic data, input the actual seismic data into the deep learning seismic lithology prediction model, and obtain actual seismic lithology prediction results.

[0041] In S1, feature extraction and seismic inversion are performed on post-stack seismic data and pre-stack gathers to obtain various seismic attributes and elastic parameter inversion results.

[0042] In S2, the geological data and logging data are combined to analyze the geological statistical characteristics hidden in them, and they are processed into a numerical format suitable for the deep neural network algorithm. First, each stratigraphic unit is represented by one-hot encoding. The six stratigraphic units are encoded separately, unit 1 is encoded as [1,0,0,0,0,0], unit 2 is encoded as [0,1,0,0,0,0], and so on. For each stratigraphic unit, the prior probability of the three lithology types (mudstone, sandstone, coal seam) is expressed as [p M ,p S ,p C ]. The 3×3 transfer matrix T is flattened into a 9-dimensional vector [T M,M ,T M,S ,T M,C ,T S,M ,T S,S ,T S,C ,T C,M ,T C,S ,T C,C ] is used to represent the transition probability between different lithologies. Among them, M, S, and C represent mudstone, sandstone, and coal seams, respectively. The thickness of mudstone and sandstone in each stratigraphic unit is divided into four different thickness categories: 1-5ms, 5-10ms, 10-20ms, and 20-70ms, and the thickness of coal seams is divided into two different thickness categories. The distribution of these thickness categories is represented as a 10-dimensional thickness distribution vector in the form of:

[0043]

[0044] In S3, multiple seismic information at the well location (such as post-stack seismic, pre-stack seismic, seismic attributes and inverted elastic parameters) is associated with a variety of geostatistical characteristics (stratigraphic units, lithology probabilities, transformation matrices and thickness distributions) through the time-depth relationship, and combined with the logging lithology characteristics constructed based on core, logging and drilling data to generate the labeled data set required for training.

[0045] In S4, based on the constructed labeled learning samples, we trained, verified and tested the deep convolutional neural network model to obtain a model constrained by multiple seismic information and multiple geological statistical characteristics. The deep neural network model is trained in a sequence-to-point manner. The input label sequence consists of multiple sampling points adjacent to the predicted point, which provides more information than single-point input. To ensure the model training effect, the appropriate time window is optimized for different branches. The cross-blind test method is adopted, that is, the blind test wells are not involved in the optimization process of the model at all. Multiple wells are used as experimental samples. Each time, a part of the wells are selected as the training data set, and the rest are used as the blind test set. The experiment is repeated to obtain the blind test results of each well. The loss function of the model uses categorical cross entropy, and the optimizer uses an optimization algorithm with adaptive learning rate and efficient convergence. In the training process, in order to prevent overfitting, an early stopping mechanism is introduced, and the model performance is further improved by dynamically adjusting the learning rate.

[0046] In S5, the trained neural network model constrained by multiple seismic information and multiple geological statistical characteristics is finally applied to the entire three-dimensional seismic data to complete the prediction of the spatial distribution of seismic lithology based on the deep neural network.

[0047] The core innovative part of the present invention mainly consists of the following parts:

[0048] 1) Coding and numerical processing of four geostatistical characteristics: The present invention converts complex geological information into numerical characteristics that can be understood and processed by deep neural networks by encoding and numerical processing of four geostatistical characteristics (stratigraphic units, lithology prior probability, lithology conversion matrix and lithology thickness distribution). Among them, stratigraphic units are represented by one-hot encoding, and lithology prior probability, lithology conversion matrix and lithology thickness distribution are processed by probabilistic numerical representation, which provides clear constraints for lithology prediction.

[0049] 2) Four geological statistical characteristics as input features of deep neural network: The present invention uses four geological statistical characteristics as input of deep convolutional neural network, so that the model can be effectively constrained by input geological prior information while using seismic data for lithology prediction. This method enhances the stability and prediction accuracy of the model, while making full use of geological background knowledge and optimizing the lithology prediction process.

[0050] 3) Synergy of multiple seismic information and multiple geostatistical feature constraints: Another innovation of the present invention is to combine multiple seismic information with multiple geostatistical feature constraints to achieve more comprehensive and accurate seismic lithology prediction. Through this synergy, the model can more accurately capture complex lithology changes, especially in the prediction of complex geological bodies such as coal seams.

[0051] The purpose of the present invention is to provide a deep learning lithology prediction method that comprehensively utilizes geological prior information and fully explores the geostatistical characteristics hidden in well logging information and geological data. Conventional deep learning-based methods use seismic data for lithofacies prediction under a data-driven supervised deep learning framework, which usually has problems such as inconsistent geological priors, poor generalization, and weak interpretability. The present invention proposes a new workflow that fully explores the geostatistical characteristics hidden in the data by fully integrating seismic data, well logging information, and geological prior knowledge, and more deeply understands the characteristics and patterns of different strata. Comprehensively utilize seismic data, well logging information, and geological prior information to fully explore the geostatistical characteristics hidden in the data and better understand the characteristics and patterns of different strata. The method uses multi-source data, including seismic data, inversion results, seismic attributes, and geostatistical laws, and designs a deep learning model that adapts to complex reservoirs in multiple strata to better capture the differences and complexity between different strata. The present invention takes into account geological prior knowledge and predicts lithofacies distribution by integrating well logging, seismic data, and geological data. Based on the laws of geological statistical characteristics, the feature set constructed by considering the horizon, lithology prior probability, lithology succession relationship, lithology thickness distribution, etc. is superior to the benchmark set composed of seismic records, inversion results and seismic attributes. It successfully achieved a more accurate simulation of the relationship between geological attributes, reflecting innovation in model consistency.

[0052] Figure 1 It is a specific process, which includes five main steps: first, attribute extraction and seismic inversion are performed; next, the geological statistical characteristics are analyzed and numerically processed; then, a reasonable lithology label sample is established based on the seismic information, lithology labels and four geological statistical characteristics at the well location; then, the deep convolutional neural network model is trained using the constructed sample labels to obtain a prediction model with multiple seismic information and multiple geological constraints; finally, the trained model is used to predict the two-dimensional or three-dimensional seismic lithology distribution.

[0053] Figure 2 It is a network architecture constrained by multiple seismic information and multiple geological statistical characteristics. The network consists of four branches: Branch 1: takes pre-stack gathers as input and provides waveform information of amplitude changes with offset distance; Branch 2: inputs time-frequency spectrum in the range of 10-80Hz and with an interval of 10Hz, which solves the problem of limited frequency bandwidth of seismic data and can predict thin layers with higher resolution; Branch 3: integrates multiple seismic attribute information; Branch 4: introduces four geological statistical characteristics (such as stratigraphic units, lithology prior probabilities, lithology conversion matrices and lithology thickness distribution, etc.).

[0054] Figure 3It is applied to a real coal-bearing clastic rock work area, and the well cross blind test results are compared and analyzed in eight groups of experiments to highlight the effectiveness of the invention; the method of the invention is applied to a work area containing coal-bearing clastic rocks, Figure 3 The results of cross-blind testing of two of the wells are shown. For comparative analysis, a total of eight groups of experiments were conducted. From right to left, they are shown: using only the inversion results as the input features of the deep neural network, the traditional threshold division method to divide the lithology; not considering the constraints of geostatistical characteristics (only using seismic pre-stack angle gathers, frequency division bodies, inversion results, and multi-seismic attributes); using only the stratigraphic constraints; using only the prior probability constraints of the lithology; using only the transformation matrix constraints; using only the thickness distribution constraints; adding all the geostatistical characteristics constraints; and the real lithology labels. Through the analysis of statistical results and prediction results, the effectiveness of the proposed lithology prediction process is proved, that is, the synergistic effect of multi-seismic information and four geostatistical characteristics constraints is better than relying solely on multi-seismic information to distinguish lithology.

[0055] Figure 4 The spatial distribution comparison of the predicted two-dimensional seismic lithology slices. Finally, the method is applied to the entire three-dimensional work area. Figure 4 The prediction results of seismic lithology slices along the horizon are shown: (a) using only multi-seismic information, and (b) using the synergistic effect of multi-seismic information and multi-geostatistical characteristic law constraints. The black circle shows the predicted lithology at the well location. Without geological constraints, it is wrongly predicted as sandstone, but the actual logging label is mudstone. The method of the present invention accurately predicts mudstone. This also further proves that the deep learning network model with geological constraints can obtain lithology prediction results that are more consistent with the sedimentary environment.

[0056] The preferred specific embodiments of the present invention are described in detail above. It should be understood that a person skilled in the art can make many modifications and changes based on the concept of the present invention without creative work. Therefore, any technical solution that can be obtained by a person skilled in the art through logical analysis, reasoning or limited experiments based on the concept of the present invention on the basis of the prior art should be within the scope of protection determined by the claims.

Claims

1. A deep learning seismic lithology prediction method based on geological statistical characteristic law constraints, characterized in that: The method comprises the following steps: S1, obtaining post-stack seismic data and pre-stack gather data, performing feature extraction and seismic inversion on the post-stack seismic data and the pre-stack gather data, and obtaining seismic attributes and elastic parameter inversion results, wherein the seismic attributes, elastic parameter inversion results, post-stack seismic data and pre-stack gather data constitute multi-seismic information; S2. Obtain geological data and well logging data, count the features of the geological data and well logging data and digitize the features, and the specific steps of digitizing the features are: uniquely encode each stratigraphic unit to obtain a coding representation, and obtain a priori probability vector of the lithology type of each stratigraphic unit, obtain a conversion probability vector between different lithologies, obtain the thickness of mudstone, sandstone and coal seam of each stratigraphic unit, and obtain a thickness distribution vector, wherein the coding represents the stratigraphic unit, the priori probability vector, the conversion probability vector and the thickness distribution vector as multiple geological statistical features; S3. Through the time-depth relationship, multiple seismic information and multiple geological statistical features are associated, corresponding lithology labels are set for the multiple seismic information, and multiple geological statistical features are used to constrain the multiple seismic information to obtain the labeled data set required for training; S4. Training, verifying and testing the deep convolutional neural network model based on the labeled data set required for training to obtain a deep learning seismic lithology prediction model; S5. Obtain actual seismic data, input the actual seismic data into the deep learning seismic lithology prediction model, and obtain actual seismic lithology prediction results.

2. A deep learning seismic lithology prediction method based on geological statistical characteristic law constraints according to claim 1, characterized in that: The encoding is expressed as: For the 6 stratigraphic units, unit 1 is encoded as [1,0,0,0,0,0], unit 2 is encoded as [0,1,0,0,0,0], unit 3 is encoded as [0,0,1,0,0,0], unit 4 is encoded as [0,0,0,1,0,0], unit 5 is encoded as [0,0,0,0,1,0], and unit 6 is encoded as [0,0,0,0,0,1].

3. The deep learning seismic lithology prediction method based on geological statistical characteristic law constraints according to claim 1 is characterized in that: The prior probability vector of lithology type is [p M ,p S ,p C ], p M ,p S ,p C Represent the prior probabilities of mudstone, sandstone and coal seams respectively.

4. The deep learning seismic lithology prediction method based on geological statistical characteristic law constraints according to claim 1 is characterized in that: The conversion probability vector between different lithologies is: [T M，M ,T M，S ,T M，C ,T S，M ,T S，S ,T S，C ,T C，M ,T C，S ,T C，C ] Among them, T M，M , T M，S , T M，C , T S，M , T S，S , T S，C , T C，M , T C，S , T C，C They respectively represent the conversion probabilities of mudstone to mudstone, mudstone to sandstone, mudstone to coal seam, sandstone to mudstone, sandstone to sandstone, sandstone to coal seam, coal seam to mudstone, coal seam to sandstone, and coal seam to coal seam.

5. The deep learning seismic lithology prediction method based on geological statistical characteristic law constraints according to claim 1 is characterized in that: The thickness distribution vector is: in They represent the distribution of mudstone 1-5ms, mudstone 5-10ms, mudstone 10-20ms, mudstone 20-70ms, sandstone 1-5ms, sandstone 5-10ms, sandstone 10-20ms, sandstone 20-70ms, They respectively represent the 1-2ms category distribution of coal seams and the 2-3ms category distribution of coal seams.

6. The deep learning seismic lithology prediction method based on geological statistical characteristic law constraints according to claim 1 is characterized in that: The deep convolutional neural network model is trained in a sequence-to-point manner, and the input label sequence consists of multiple sampling points adjacent to the point to be predicted.

7. The deep learning seismic lithology prediction method based on geological statistical characteristic law constraints according to claim 1 is characterized in that: The deep convolutional neural network model is trained using a cross blind testing method.

8. The deep learning seismic lithology prediction method based on geological statistical characteristic law constraints according to claim 7 is characterized in that: The specific steps of training using the cross blind test method are: The labeled data set is divided into training well data and a blind well logging data. The training well data is input into the deep convolutional neural network model for training verification, and the blind well logging data is input into the trained deep convolutional neural network model for testing. The above steps are repeated to obtain the blind test results of each well. Based on the blind test results of each well, the well data with prediction accuracy less than the threshold are screened, and the screened well data are input into the deep convolutional neural network model for training again.

9. The deep learning seismic lithology prediction method based on geological statistical characteristic law constraints according to claim 1 is characterized in that: The early stopping mechanism is adopted in the training process of the deep convolutional neural network model.

10. The deep learning seismic lithology prediction method based on geological statistical characteristic law constraints according to claim 1 is characterized in that: The learning rate is dynamically adjusted during the training process of the deep convolutional neural network model.

Images